Approach Advancing Stock Market Forecasting with Joint RMSE Loss LSTM-CNN Model DOI
Mungara Kiran Kumar, Jagdish Chandra Patni, Mohan Raparthi

et al.

Fluctuation and Noise Letters, Journal Year: 2023, Volume and Issue: 23(02)

Published: Dec. 17, 2023

The intricacies and dynamism of financial markets pose challenges to models seeking comprehensively capture the multitude factors influencing stock price movements. As such, there remains room for improvement in forecasting accuracy. In response, we introduce a novel approach that unifies Root Mean Square Error (RMSE), loss functions Long Short-Term Memory (LSTM) Convolutional Neural Networks (CNN). By concurrently optimizing their RMSE functions, our takes use capabilities LSTM learning long-term time series relationships CNN extracting deep features from data. To maximize efficacy each model branch within this unified framework, split training set into two different representations, one consisting standard data other picture We compare proposed others field demonstrate its viability, particularly Backpropagation (BP), LSTM, CNN, fusion LSTM-CNN model. Experimental evaluations conducted on three diverse datasets—Development Bank, Stock Connect Index (SCI), Composite (CI)—validate robust predictive performance applicability joint model, thus showcasing potential forecasting.

Language: Английский

Using Deep Learning to Identify Deepfakes Created Using Generative Adversarial Networks DOI Creative Commons

Jhanvi Jheelan,

Sameerchand Pudaruth

Computers, Journal Year: 2025, Volume and Issue: 14(2), P. 60 - 60

Published: Feb. 10, 2025

Generative adversarial networks (GANs) have revolutionised various fields by creating highly realistic images, videos, and audio, thus enhancing applications such as video game development data augmentation. However, this technology has also given rise to deepfakes, which pose serious challenges due their potential create deceptive content. Thousands of media reports informed us occurrences, highlighting the urgent need for reliable detection methods. This study addresses issue developing a deep learning (DL) model capable distinguishing between real fake face images generated StyleGAN. Using subset 140K dataset, we explored five different models: custom CNN, ResNet50, DenseNet121, MobileNet, InceptionV3. We leveraged pre-trained models utilise robust feature extraction computational efficiency, are essential features. Through extensive experimentation with dataset sizes, preprocessing techniques, split ratios, identified optimal ones. The 20k_gan_8_1_1 produced best results, MobileNet achieving test accuracy 98.5%, followed InceptionV3 at 98.0%, DenseNet121 97.3%, ResNet50 96.1%, CNN 86.2%. All these were trained on only 16,000 validated tested 2000 each. was built simpler architecture two convolutional layers and, hence, lagged in its limited capabilities compared deeper networks. research work included user-friendly web interface that allows deepfake uploading images. backend developed using Flask, enabling real-time detection, allowing users upload analysis demonstrating practical use platforms quick, verification. application demonstrates significant applications, social platforms, where can help prevent spread content flagging suspicious review. makes important contributions comparing models, including understand balance complexity detection. It identifies setup improves while keeping costs low. Additionally, it introduces tool making useful moderation, security, Nevertheless, identifying specific features GAN-generated deepfakes remains challenging high realism. Future works will aim expand all 140,000 refine increase accuracy, incorporate more advanced Vision Transformers diffusion models. outcomes contribute ongoing efforts counteract negative impacts

Language: Английский

Citations

1
Deep learning-based biometric image feature extraction for securing medical images through data hiding and joint encryption–compression DOI
Monu Singh, Naman Baranwal,

Kedar Nath Singh

et al.

Journal of Information Security and Applications, Journal Year: 2023, Volume and Issue: 79, P. 103628 - 103628

Published: Oct. 29, 2023

Language: Английский

Citations

11
Performance analysis of state‐of‐the‐art CNN architectures for brain tumour detection DOI Open Access
Hafiz Muhammad Tayyab Khushi, Tehreem Masood, Arfan Jaffar

et al.

International Journal of Imaging Systems and Technology, Journal Year: 2023, Volume and Issue: 34(1)

Published: Aug. 18, 2023

Abstract Deep learning models, such as convolutional neural network (CNN), are popular now a day to solve various complex problems in medical and other fields, image classification, object detection, recommendation of images, processing natural languages video analysis. So, the idea studying architecture CNNs has gotten lot attention become popular. This study analysed contrasted performance many different CNN models trained on publicly accessible Br35h dataset for detection brain tumours. These included LeNet, AlexNet, VGG16, VGG19 ResNet50. Several optimisers were used this research fine‐tune model. Adam (adaptive moment estimation), SGD (stochastic gradient descent) RMSprop (root‐mean‐square propagation). Accuracy, miss‐classification rate, sensitivity, specificity, NPV (negative predictive value), PPV (positive F1‐score false omission rate (FOR) assess efficacy five architectures using three optimisers. The experimental results showed that AlexNet with optimiser performed better than achieved highest accuracy 98.79% miss classification 1.20%. It also 98.98% 98.58% 98.93% NPV, 98.65% PPV, 98.82% 1.06% FOR.

Language: Английский

Citations

10
A Novel Hybrid Framework for Deepfake Detection DOI

Samarth Dhol,

Nishant Kanani,

Diya Koyani

et al.

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: April 17, 2025

Abstract Fast developments in artificial intelligence technology produced out- standing generative advancements that highly realistic deepfake media content which consistently outpaces existing method detection capabilities. The rising distribution of synthetic leads to urgent threats for authen- ticity because privacy and security issues remain even though systems need be implemented immediately. proposed combination lightweight model solves present problems related spatial-temporal fea- ture analysis adaptive adversarial noise reduction noise-resilient feature extraction methods.The Xception backbone operates with temporal attention find inconsistencies between compressed video frames through the Celeb-DF-v1 Celeb-DF-v2 datasets at real-time speeds inference. Deepfake de- tection on achieved a top-tier success rate 90% accuracy thus surpassing all current competing solutions by 5.7%. At same time pro- posed maintained strong effectiveness when facing actual defor- mation challenges different dataset environments. shows great efficiency adaptability making it ideal social en- vironments where defends against evolving dangers scale. Supported future enhancement research we consider limitations im- proving attacks previously unobserved adversary conditions.

Language: Английский

Citations

0
Cyber Social Threats: A Data-centric AI Perspective DOI
Adita Kulkarni

Published: April 24, 2025

Language: Английский

Citations

0
Enhancing Fake Image Detection with Ensembled Convolutional Neural Networks DOI

Adeeb Khan,

Sarsij Tripathi

Published: May 13, 2025

Abstract Fake image detection has emerged as a vital task for the Generative AI era due to fast evolution in generations of models that have made highly realistic synthetic images possible. In this paper, we formulate an ensemble-based Convolutional Neural Network (CNN) enhance fake accuracy. Our methodology includes training five CNN on separate datasets consisting real and artificially created found different public datasets. The are produced using latest include StyleGAN2, StyleGAN3, Diffusion GAN, Taming Transformer Gansformer. outputs fused stacking ensemble process which several classifiers such Random Forest, Gradient Boosting, AdaBoost, Support Vector Machine, Multi-Layer Perceptron Logistic Regression utilized boost final classification performance. ultimate test unseen data reveals increase performance our approach exhibits high accuracy rate more than 90%. Comparison metrics precision, recall F1-score complete insight about proposed approach. These results indicate use deep learning approaches makes systems strongly robust nature even applicable real-world settings.

Language: Английский

Citations

0
Image Classification and Detection of Artificial Images Using CNN Models DOI

Rajkumar Devendran,

A S Aneetha

Communications in computer and information science, Journal Year: 2025, Volume and Issue: unknown, P. 79 - 88

Published: Jan. 1, 2025

Language: Английский

Citations

0
A flexible analytic wavelet transform and ensemble bagged tree model for electroencephalogram-based meditative mind-wandering detection DOI Creative Commons

Ajay Dadhich,

Jaideep Patel,

Rovin Tiwari

et al.

Healthcare Analytics, Journal Year: 2023, Volume and Issue: 5, P. 100286 - 100286

Published: Dec. 4, 2023

Mind-wandering (MW) is when an individual's concentration drifts away from the task or activity. Researchers found a greater variability in electroencephalogram (EEG) signals due to MW. Collecting more nuanced information raw EEG data examine harmful effects of MW time-consuming. This study proposes multi-resolution assessment using flexible analytic wavelet transform (FAWT). The FAWT algorithm decomposes into representative sub-bands (SBs). Several statistical characteristics are derived obtained SBs, and during meditation on investigated. A set significant chosen fed machine learning modules 10-fold validation approach detect subjects automatically. Our proposed framework attained highest classification accuracy 92.41%, sensitivity 93.56%, specificity 91.97%. can be used design suitable brain-computer interface (BCI) system reduce increase depth for holistic long-term health society.

Language: Английский

Citations

6
Domain-invariant and Patch-discriminative Feature Learning for General Deepfake Detection DOI
Jian Zhang, Jiangqun Ni, Fan Nie

et al.

ACM Transactions on Multimedia Computing Communications and Applications, Journal Year: 2024, Volume and Issue: unknown

Published: April 27, 2024

Hyper-realistic avatars in the metaverse have already raised security concerns about deepfake techniques, deepfakes involving generated video “recording” may be mistaken for a real recording of people it depicts. As result, detection has drawn considerable attention multimedia forensic community. Though existing methods achieve fairly good performance under intra-dataset scenario, many them gain unsatisfying results case cross-dataset testing with more practical value, where forged faces training and datasets are from different domains. To tackle this issue, paper, we propose novel Domain-Invariant Patch-Discriminative feature learning framework - DI&PD. For image-level learning, single-side adversarial domain generalization is introduced to eliminate variances learn domain-invariant features samples manipulation methods, along global local random crop augmentation strategy generate data views images at various scales. A graph structure then built by splitting learned maps, each spatial location corresponding patch, which facilitates patch representation message-passing among similar nodes. Two types center losses utilized discriminative both patch-level embedding spaces. Extensive experimental on several demonstrate effectiveness proposed method compared other state-of-the-art methods.

Language: Английский

Citations

2
Convolutional Neural Networks for Automated Diagnosis of Diabetic Retinopathy in Fundus Images DOI Creative Commons
S. Rama Krishna, Naresh Cherukuri,

Y Dileep Kumar

et al.

Journal of Artificial Intelligence and Technology, Journal Year: 2023, Volume and Issue: unknown

Published: Aug. 24, 2023

Diabetic retinopathy (DR), a long-term complication of diabetes, is notoriously hard to detect in its early stages due the fact that it only shows subset symptoms. Standard diagnostic procedures for DR now include OCT and digital fundus imaging. If images alone could provide reliable diagnosis, then eliminating costly optical coherence tomography would be beneficial all parties involved. Optometrists their patients will find this useful. Using deep convolutional neural networks, we novel approach problem. Our deviates from standard DCNN methods by exchanging typical max-pooling layers with fractional ones. In order collect more subtle information categorisation, two such DCNNs, each different number layers, are trained. To establish these limits, use networks (DCNNs) features extracted picture metadata train support vector machine classifier. our experiments, used Kaggle's open detection database. We fed model 34,124 training images, 1,000 validation examples, 53,572 test it. Each five classes proposed classifier corresponds one steps process given numeric value between 0 4. Experimental results show higher identification rate (86.17%) than those found existing literature, indicating suggested strategy may effective. have jointly developed an algorithm learning accompanying software, we've named Deep Retina. Images acquired person using portable ophthalmoscope instantly analyzed technology. This technology might self-diagnosis, at-home care, telemedicine.

Language: Английский

Citations

4